Even for high frequency variants, neuroimaging databases of 1,000 subjects would be underpowered to detect commonly observed effect sizes. We must bear in mind that, in the standard experimental design (see below for others), the significance of a genome-wide hit has to be at least 20 million to one, to account for the very large number of variants tested. Some have argued that neuroimaging studies reporting effects of candidate genes—such as COMT or BDNF—are also at risk for false positive effects, in the sense that any number of genes could have been assessed, with no way to verify whether the paper was selectively reporting the successes (Flint and Munafo 2013; Ioannidis 2005). While this problem is shared by selective reporting of results in many fields, imaging genetics is particularly at risk because of the ease of retesting the same data. Some have raised the concern that a considerable proportion of neuroimaging genetics associations, especially those found in small samples, may not replicate in subsequent analysis. Clearly, although the number of genes in the human genome is limited, an almost unlimited number of candidate genes could be still tested (Bishop 2013; Flint and Munafo 2013).
